ReSPARQL: a SPARQL Extension for Generic Recommendations on RDF-graphs

The abundance of data published using Semantic Web technologies ratifies their high degree of maturity reached. Moreover, the flexibility of the Resource Description Framework (RDF) enables it to model any knowledge within a specific domain. This has given rise to a potential use of RDF data as input for applications which were not originally designed to operate online on Web data sources. Recommender systems are one example of such applications. These aim to predict the taste of a user towards a set of not consumed items and are typically well optimized for fixed domains. The benefit of having a recommender system which takes advantage of Web knowledge is that a user could be assisted in selecting information from the Web and, therefore, reducing the information overload. However, these systems cannot handle the diversity and unstructuredness of Semantic Web data. One of the reasons is that Semantic Web query languages, such as SPARQL, support retrieval of data exclusively based on facts; predictions or suggestions are entities that cannot be explicitly retrieved. In this thesis ReSPARQL will be presented: an extension of the SPARQL syntax and semantics that fills this gap and enables a generic and flexible approach for recommendations over arbitrary RDF-graphs. It supports content-based and collaborative filtering recommendations and allows both paradigms to gain benefit from each other.